Geopolitics
Trump and Xi Built an AI Hotline. Nobody Defined 'Incident.'
The Trump-Xi summit produced a Super Intelligence dialogue and an AI incident hotline — one sentence, no date, and no definition of 'incident.'
AI
Technology
Intel Can Fill Half Its CPU Orders. Memory Gets Worse in 2027.
Intel's CEO says it can meet half of CPU demand and the memory shortage worsens next year. SK hynix sees it lasting to 2030. Budget your 2027 refresh now.
AI
Trump Renames AI at the UN. A Day Later, Its Builders Ask for Rules.
Trump told the UN that AI is now 'super intelligence' and rejected a 'globalist scheme' to control it. A day later, Altman and Amodei asked for rules.
AI
GPT-6 Sol Arrives at Half the Price
OpenAI's new coding model costs half as much as GPT-5.6 Sol. Independent tests find similar overall ability, uneven gains, and reasons to check the work.
AI
Claude Opus 5.5 Takes the Top Spot and Costs Less Than Opus 5
Anthropic's Opus 5.5 leads an independent index, costs 20% less per token than Opus 5, and is the first Opus to launch with Claude's text watermark.
AI
Grok 4.7 Keeps the $2 Price. The Tasks Got Longer.
SpaceXAI released Grok 4.7 on September 21 at Grok 4.6's $2 and $6 rates. Coding scores rose, and outside tests show each finished task using far more tokens.
AI
Both Parties Are Running Against Data Centers. Trump Isn't.
Candidates from both parties are weaponizing the data center backlash in the midterms. The president is nearly alone on the other side.
Healthcare IT
FDA Won't Let Radiology AI Skip Premarket Review. Here's Why.
FDA formally denied Harrison.ai's bid to let experienced radiology AI vendors skip 510(k) review for new tools. Why regulators and radiologists said no.
AI
Claude Now Leads 26% of Anthropic's Research, by Anthropic's Count
Anthropic's new index says Claude leads 26 percent of its AI research, up from under 1 percent in February. The number is real, self-measured, and unverified.
AI
Microsoft's AI Code: Never Resist Shutdown, Never Hide Your Reasoning
Microsoft's draft code for its own models bars resisting shutdown, own goals, or hidden reasoning. Six weeks of comment, then it shapes the 2027 models.